American Journal of Botany 98(1): 76–87. 2011.

O LD – NEW WORLD AND TRANS-AFRICAN DISJUNCTIONS OF THAMNOSMA (RUTACEAE): INTERCONTINENTAL LONG-DISTANCE DISPERSAL AND LOCAL DIFFERENTIATION IN THE SUCCULENT BIOME 1

Mike Thiv2,7 , Timothe ü s van der Niet3 , Frank Rutschmann4 , Mats Thulin5 , Thomas Brune6 , and Hans Peter Linder4

2 Botany Department, State Museum of Natural History Stuttgart, Rosenstein 1, 70191 Stuttgart, Germany; 3 School of Biological and Conservation Sciences, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209, South Africa; 4 Institute for Systematic Botany, University of Zurich, Zollikerstrasse 107, 8008 Zurich, Switzerland; 5 Department of Systematic Biology, Uppsala University, Norbyv ä gen 18D 75236 Uppsala, Sweden; and 6 Institute of Food Science and Biotechnology, University of Hohenheim, Garbenstrasse 25, 70593 Stuttgart, Germany

• Premise of the study : The succulent biome is highly fragmented throughout the Old and New World. The resulting disjunctions on global and regional scales have been explained by various hypotheses. To evaluate these, we used Thamnosma , which is restricted to the succulent biome and has trans-Atlantic and trans-African disjunctions. Its three main distribution centers are in southern North America, southern and eastern Africa including Socotra. • Methods: We conducted parsimony, maximum likelihood, and Bayesian phylogenetic analyses based on chloroplast and nu- clear sequence data. We applied molecular clock calculations using the programs BEAST and MULTIDIVTIME and biogeo- graphic reconstructions using S-DIVA and Lagrange. • Key results: Our data indicate a weakly supported paraphyly of the New World species with respect to a palaeotropical lineage, which is further subdivided into a southern African and a Horn of Africa group. The disjunctions in Thamnosma are mostly dated to the Miocene. • Conclusions : We conclude that the Old – New World disjunction of Thamnosma is likely the result of long-distance dispersal. The Miocene closure of the arid corridor between southern and eastern Africa may have caused the split within the Old World lineage, thus making a vicariance explanation feasible. The colonization of Socotra is also due to long-distance dispersal. All recent Thamnosma species are part of the succulent biome, and the North American species may have been members of the arid Neogene Madro-Tertiary Geofl ora. Phylogenetic niche conservatism, rare long-distance dispersal, and local differentiation account for the diversity among species of Thamnosma .

Key words: biogeography; disjunctions; Madro-Tertiary fl ora; molecular dating; phylogeny; Rutaceae; succulent biome; Thamnosma .

Schrire et al. (2005b) accounted for the diversifi cation of le- netic niche conservatism, widely distributed clades should occupy gumes with a model that incorporated phylogenetic niche con- similar habitats in remote areas, a pattern neatly demonstrated servatism with local differentiation. The tendency of lineages to in the legumes (Lavin et al., 2004; Schrire et al., 2005b). The retain their ancestral ecological niche rather than adapting to succulent biome, which is characterized by a climate with a new environments (phylogenetic niche conservatism) has has long dry season, is found in relatively small fragments and been demonstrated in diverse clades and regions ( Wiens, 2004 ; isolated regions in both South and North America, in a large Donoghue, 2008 ; Crisp et al., 2009 ). On the basis of phyloge- region in SW Africa, and from Tanzania to Pakistan. The pre- dominant growth forms in these climates are usually evergreen, sclerophyllous, nonfi re-adapted shrubs and succulents ( Mooney 1 Manuscript received 6 September 2010; revision accepted 23 November 2010. and Dunn, 1970; Schrire et al., 2005a). This succulent biome The authors would like to thank Jon Rebman (SDSU, USA) and Michael has been the object of some studies (e.g., Lavin et al., 2004 ; Denslow (Rancho Santa Ana Botanic Garden, USA) for providing plant Schrire et al., 2005a, b , 2009 ). Its characteristic climate and material, Joachim Kadereit (Mainz, Germany) for helpful comments on the highly fragmented distribution make it an excellent system in manuscript, and Mohamed Ali Hubaishan (AREA Research Station which to investigate the effects of niche conservatism. Mukalla), Ahmed Said Sulaiman (EPA Socotra), Said Masood Awad The succulent biome shows a disjunction between the New Al-Gareiri (Dept. Agriculture Socotra), and Mohamed El-Mashjary (EPA and the Old World. This pattern is matched by the distribution Sanaa, all Yemen) for support of the fi eldwork in Socotra. The fi eldwork of legumes where, e.g., the genera Arcoa Urb., Diphysa Jaqu., was conducted as part of the BIOTA Yemen Project funded by the German Pictetia DC., and Chapmannia Torr. & Gray occur in North and Ministry for Research and Education (BMBF). The work was supported by grants of the German Research Foundation (DFG, Th830/1-1) and the Mesoamerica and Tetrapterocarpon Humbert, Zygocarpum Claraz-Schenkung (Switzerland) to M.T. Thulin & Lavin, Ormocarpopsis R.Vig., Ormocarpum P. Beauv., 7 Author for correspondence (e-mail: [email protected]) and Chapmannia occur in Africa. In most of these cases, the New World group fi nds its corresponding sister clade in Africa. doi:10.3732/ajb.1000339 Several hypotheses were assumed to explain this pattern. Inferring

American Journal of Botany 98(1): 76–87, 2011; http://www.amjbot.org/ © 2011 Botanical Society of America 76 January 2011] Thiv et al. —T HAMNOSMA 77 from physiognomical similarities, links have been proposed where three species occur in the southern continent from An- between succulent biome taxa and dry boreotropical and gola to the northern regions of South Africa, and another three Madrean-Tethyan representatives (Lavin et al., 2000). Based species grow around the Horn of Africa extending to southern on palaeontological and geological evidence, Tiffney (1985b) Arabia and the island of Socotra ( Thulin, 1999 ) ( Table 1 , Fig. 1 ). proposed an Eocene North Atlantic Land bridge (55.8 – 33.9 This pattern with trans-Atlantic and trans-African disjunctions million years ago [Ma]) linking the continents of the Northern of the succulent biome makes Thamnosma a suitable case study Hemisphere, resulting in a circumboreal, deciduous, and ever- (1) to test the hypotheses on the origin of Old and New World green heterogenous fl ora (boreotropical hypothesis, Wolfe, disjunctions in the succulent biome, (2) to investigate the “ arid 1975). Lavin et al. (2000) suggested that vicariance between corridor ” between southern Africa and the Horn of Africa North/Mesoamerica and Africa was possible via this land bridge region, and (3) to evaluate a vicariance origin for the Socotran for legumes. Similarily, only younger in time, the Madrean- taxon. Tethyan hypothesis suggested a common occurrence of recent and fossil sclerophyllous taxa in both southwestern North America and the Old World by the late Eocene to the late Paleo- MATERIALS AND METHODS gene (37.2 – 23 Ma, Axelrod, 1972 , 1973 , 1975 ). Another pos- sible vicariant origin of this disjunction may have been caused Taxon sampling— To resolve the interspecifi c phylogenetic relationships by an early Miocene (23– 16 Ma) migration route via the Bering of Thamnosma, we included all 11 species of Thamnosma (Johnston, 1983; land bridge, as suggested by Stebbins and Day (1967) and Tiff- Thulin, 1999) in a combined analysis of the ITS, matK , and the atpB-rbcL spacer ( Table 2 ), rooted with the the closely related Ruta L. (Salvo et al., ney (1985a). In contrast to these vicariance scenarios, the inter- 2008). On the basis of three chloroplast markers, Thamnosma is closely related continental disjunction of legumes in the succulent biome was to the eastern Asian, monotypic Boenninghausenia Reichb. ex Meissner, Ruta recently interpreted as result of long distance because the ages L., Haplophyllum A. Juss., and Cneoridium dumosum Hook. f. ( Salvo et al., of these lineages are much younger than the assumed vicariance 2008). Of those, only Cneoridium Hook. f. and Ruta could be included in our events ( Lavin et al., 2004 ). study due to availability of material. Salvo et al. (2008) found a sister-group The succulent biome is also disjunct on the African conti- relationship between Thamnosma and Boenninghausenia . Still, the monotypic Psilopeganum Hemsl. ex Forb. & Hemsl., restricted to a small area of central nent. Numerous plants (e.g., Aizoon L. [Aizoaceae], Trichoneura China (Song et al., 2004), may also be closely related to Thamnosma , because Anderss. [Poaceae]) and animals (e.g., ostriches) occupy an these are the only bicarpellate genera in the Rutinae and share capitate stigmas “ arid corridor ” between southwestern and northeastern Africa and similar fl ower merosity and seed shapes (Engler, 1897, 1931 ; Salvo et al., ( Balinsky, 1962 ; Verdcourt, 1969 ; de Winter, 1971 ; J ürgens, 2008 ). This relationship is also supported by phytochemical data (acridones of 1997 ), with an extension to Mauretania (NW Africa). The clo- type H1; da Silva et al., 1988). Because Psilopeganum shares more morpho- sure and widening of this arid track has been linked to Miocene logical synapomorphies with Thamnosma rather than with Boenninghausenia (cf. Salvo et al., 2008 ), it seems more likely that Thamnosma and Psilopega- uplift and rifting in Central Africa and climate changes in the num are phylogenetic sisters. However, it is unlikely that Psilopeganum is Pleistocene ( Caujape-Castells et al., 2001 ; Coleman et al., embedded in Thamnosma because they differ in the degree of carpel fusion 2003), which have led to isolated populations in these areas and the size of their disc (Engler, 1897, 1931 ). No DNA is available for Psi- characterized by a short rainy season. Evidence supporting the lopeganum ; thus these morphologically based interpretations have not yet arid track came from phylogeographical studies of ostriches been tested. Material was mostly taken from herbarium specimens or, in some (Freitag and Robinson, 1993) and phylogenetic investigations cases, silica-gel-dried leaves. Vouchers and EMBL accession numbers are given Table 2 . of Androcymbium Willd. (Colchicaceae, Caujape-Castells et al., To estimate the ages of biogeographically relevant nodes in the Thamnosma 2001 ), Senecio L. ( Coleman et al., 2003 ), and Zygophyllum phylogeny, we assembled a set of species representing all major clades of Ru- L. (Zygophyllaceae, Bellstedt et al., 2008 ). taceae and two taxa with reliable fossils for the rbcL analysis. Thamnosma was A small transoceanic disjunction in the succulent biome is found in the Horn of Africa region. This area covers the three provinces of the Eritreo– Arabian subregion of Takhtajan Table 1. Species of Thamnosma , their distribution and defi ned areas of (1986) : Somalo-Ethiopia, South Arabia, and the Socotran ar- endemism: H = Horn of Africa region, N = Northern/ Mesoamerica, chipelago. Socotra is of continental origin and separated from S = Southern Africa. the Arabian plate about 15– 18 Ma (Richardson et al., 1995; Area of Fleitmann et al., 2004; Van Damme, 2009). In all three regions, Taxon Species distribution endemism the succulent biome is prevalent, leading to two explanations for Socotran – continental disjunctions. They are either vicariant Thamnosma africana Engl. S Angola, Namibia S relicts preceeding the rifting (postulated for chameleons; Macey Thamnosma crenata (Engl.) South Africa, Northern Province S Thulin et al., 2008 ), or were established by long-distance dispersal Thamnosma hirschii Stapf Yemen, Oman, N Somalia, H over water (e.g., Aerva Forssk. [Amaranthaceae], Thiv et al. Socotra [2006 ]; Echidnopsis Hook. f. [Apocynaceae-Asclepiadoideae], Thamnosma montana Torr. & S USA, Mexico N Thiv and Meve [2007]; and Campylanthus Roth [Plantag- Fr é m. inaceae], Thiv et al. [2010] ). Thamnosma pailensis Mexico, SE Coahuila N To explore the evolution of these disjunctions, we used the M. C. Johnst. Thamnosma rhodesica S+W Zimbabwe, Botswana S rutaceous genus Thamnosma Torr. & Fr é m. With only 11 spe- (Bak. f.) Mendon ç a cies, Thamnosma occurs in three widely disjunct centers of dis- Thamnosma socotrana Balf. f. Yemen, Socotra H tribution. Five species grow in southwestern North America, Thamnosma somalensis Thulin NE Somalia H these could well be relicts of the North American Madro- Thamnosma stanfordii Mexico, S Coahuila N Tertiary geofl ora (Axelrod, 1958), an assemblage of sclerophyl- I. M. Johnst. lous and microphyllous taxa that were adapted to conditions of Thamnosma texana Torr. S USA, Mexico N Thamnosma trifoliata Mexico, Baja California N low annual precipitation, high summer temperatures, and long I. M. Johnst. sunshine periods. The other species are mainly found in Africa, 78 American Journal of Botany [Vol. 98

Fig. 1. Distribution map (http://www.aquarius.ifm-geomar.de) of Thamnosma in North and Central America and in Africa.

represented by six North American and Old World species: southern African products were analyzed using automated sequencing systems ABI PRISM 3100 Thamnosma species were not included because amplifi cations did not succeed, (PE Biosystems, Darmstadt, Germany). possibly from degradation of the DNA extracted from herbarium material. Due to the use of degraded DNA from herbarium material, phylogenetic reconstruc- Phylogenetic analysis— Sequences of all Thamnosma species were aligned tion and molecular dating were based on only half of the gene length of rbcL using the program Clustal X version 1.81 (Thompson et al., 1997) and then (790– 813 bp). manually adjusted. The alignments are available in TreeBase (S10787 , http:// treebase.org). First, parsimony (MP) bootstrap analyses (options given later) Laboratory work— The molecular work followed standard protocols. For were conducted for both chloroplast and nuclear data. To test for congruence DNA extraction, the DNeasy plant extraction kit (Qiagen) was used according between these two data sets, we conducted a partition homogeneity test as im- to the manufacturer’ s protocol. Amplifi cations were performed using 1.5 mM plemented in the program PAUP* 4.0b10 ( Swofford, 1998 ). Because this test   buffer, 0.625 mM MgCl 2 , 0.2 mM dNTPs, 0.05 U/ L Taq DNA polymerase failed to reject the hypothesis of congruency, the two data sets were combined (Amersham Biosciences, Freiburg, Germany), 0.325  M primer, and 5 ng/  L ( Johnson and Soltis, 1998 ; Wiens, 1998 ). Maximum likelihood (ML; Felsenstein, DNA template. PCR profi les included 33 cycles of 94 C for 1 min, 50– 55  C for 1981 ) and parsimony analyses for the combined data were performed using 1 min, and 72  C for 2– 3 min. For amplifi cations and sequencing, the following PAUP* 4.0b10. For the ML analysis, the GTR+I+   model was used as indi- primers were used. ITS nrDNA: ITS-A 5a -GGAAGGAGAAGTCGTAA-  cated by the Akaike information criterion (AIC) in the program Modeltest 3.06 CAAGG-3a , ITS-B 5a -CTTTTCCTCCGCTTATTGATATG-3a , ITS-D 5a -CTC- ( Posada and Crandall, 1998 ). For parsimony analyses, characters were equally TCGGCAACGGATATCTCG-3a  (all Blattner, 1999), ITS-R2 5a -CGTTC-  weighted, character states were treated as unordered, and indels were treated as AAAGACTCGATGGTTC-3a (von Balthazar et al., 2000). Multiple DNA missing data. Parsimony analyses were carried out using 103 random-addition- extractions and ITS sequencing for several accessions of Thamnosma , yielded sequence replicates, with Multrees in effect and tree- bisection-reconnection identical results indicating the absence of paralogous copies of ITS. The atpB - (TBR) swapping. Parsimony bootstrap analyses (104 replicates) were calculated rbcL intergenic spacer: atpB-F2 5a -GAAGTAGTAGGATTGATTCTC-3  a , using the closest Multrees and TBR options. Additionally, Bayesian inference atpB-R5 5a -GAAGTAGTAGGATTGATTCTC-3 a  (both Manen et al., 1994 ); using the combined data set was explored using the program MrBayes v. 3.1.2 matK (these primers amplify the 3 a end of the matK gene and the adjacent ( Huelsenbeck and Ronquist, 2001 ). The GTR+  model was selected by AIC in intron): matK-Th-F 5a -TTATTCATCTGATTGGATCAT-3 a  (designed in the the program MrModeltest (Nylander, 2004). Two runs using four parallel present study), matK-1R 5a -GAACTAGTCGGATGGAGTAG-3 a ( Sang et al., chains of 5   10 6 generations with heats of 1.00, 0.83, 0.71, and 0.62 were 1997); rbcL: rbcL-F1 5a -TCACCACAAACAGARACTAAAGC-3 a , rbcL-R2 performed, with a sample frequency of 102 . Trees from the fi rst 104 generations 5 a -RCGRTGRATGTGAAGAAG-3 a (both Long-Yin Qiu, University of Michi- were discarded. gan, personal communication). The use of herbarium material only allowed the We tested alternative scenarios using a parametric bootstrap analysis. amplifi cation of a partition of rbcL . PCR products were cleaned using the PCR This test is shown to be a statistically sound method of evaluating different purifi cation kit (Qiagen, Hilden, Germany). Cycle sequencing was conducted alternative topological hypotheses (Huelsenbeck et al., 1996; Goldman et al., using ABI PRISM BigDye 2.1 to obtain double stranded sequences. Resulting 2000; Stefanovi 4 and Olmstead, 2004 ). This procedure included the determination January 2011] Thiv et al. —T HAMNOSMA 79

Table 2. Taxon sampling, vouchers and new EMBL numbers for this study. Herbaria acronyms are according to Index Herbariorum (http://sciweb.nybg. org/science2/IndexHerbariorum.asp).

atpB-rbcL Taxon Collector No. Origin voucher ITS1 ITS2matK spacerrbcL

Cneoridium dumosum Thibault et s.n. Cult. Rancho Santa Ana RSA FN552678 Hook.f. Denslow Botanic Garden RSA674337 Euodia hupehensis Dode Wagen s.n. Cult. Bot. Garden Zurich Z FN552679 Ruta chalepensis L. Wagen s.n. Cult. Bot. Garden Zurich Z FN552647 FN552663 FN552615 FN552631 Thamnosma africana Goldblatt et al. 8927 Namibia MO FN552649 FN552665 FN552617 FN552633 Thamnosma africana Lavranos 21900 Namibia MO FN552648 FN552664 FN552616 FN552632 Thamnosma crenata Venter 11224 South Africa MO FN552650 FN552666 FN552618 FN552634 Thamnosma hirschii Miller 10000 Yemen, Socotra B FN552651 FN552667 FN552619 FN552635 Thamnosma hirschii Kilian & Hein NK 6164 Yemen B FN552653 FN552669 FN552621 FN552637 Thamnosma hirschii Thiv 3187 Yemen, Socotra Z FN552652 FN552668 FN552620 FN552636 FN552680 Thamnosma montana Landrum et 9022 USA, Arizona NY FN552654 FN552670 FN552622 FN552638 FN552681 Landrum Thamnosma montana White 4252 USA, California BM FN552655 — FN552623 FN552639 FN552682 Thamnosma pailensis Woodruff 369 Mexico, Coahuila BM FN552656 FN552671 FN552624 FN552640 FN552683 Thamnosma rhodesica Blomberg et al. BMP 104 Botswana UPS FN552657 FN552672 FN552625 FN552641 Thamnosma socotrana Thiv 3176 Yemen, Socotra STU, Z FN552658 FN552673 FN552626 FN552642 Thamnosma socotrana Kilian 2495 Yemen, Socotra B FN552684 Thamnosma somalensis Thulin et al. 9489 Somalia UPS FN552659 FN552674 FN552627 FN552643 Thamnosma stanfordii Chiang et al. 9545 Mexico, Coahuila MO, NY FN552660 FN552675 FN552628 FN552644 Thamnosma texana McGolderick s.n. USA, Texas L FN552661 FN552676 FN552629 FN552645 FN552685 Thamnosma trifoliata Rebman 7577 Mexico, Baja California SD FN552662 FN552677 FN552630 FN552646 FN552686 of ML parameters for the described constrained topologies. Based on these pa- procedure with the following settings for the prior distributions: 1.50 for both rameters, 99 simulated data sets were created using the program Seq-Gen rttm and rttmsd , 0.07 for both rtrate and rtratesd , 0.4 for both brownmean and (Rambaut and Grassly, 1997). The simulated data sets were analyzed using brownsd , and 84 million years ago (Ma) for bigtime , which is the maximum age maximum parsimony with the closest sequence addition and TBR branch swap- of Sapindales according to Wikstr ö m et al. (2001). This maximum age means ping, testing for signifi cant differences in lengths between the constrained tree that this node cannot be estimated to be older than this date. We ran the Markov as null hypothesis and the optimal tree. chain for at least 103 cycles and collected one sample every 102 cycles, after an unsampled burn-in of 10 4 cycles. We repeated the analyses in BEAST and Molecular dating —We used a molecular clock to date the disjunction be- MULTIDIVTIME twice using different random starting number to assure con- tween New and Old World species of Thamnosma . Based on a likelihood ratio vergence of the Markov chain and combined the results. (LR) test (Felsenstein, 1981; Sanderson, 1998; Nei and Kumar, 2000), substitu- The rate-corrected tree was calibrated with two fossil taxa. A continuous tion rates of the rbcL sequences were near clock-like. Nonetheless, we em- macrofossil record starting form the Late Eocene/Early Oligocene (40 – 35 Ma) ployed Bayesian dating using a relaxed clock. This method (Thorne et al., 1998; is available for Euodia J.R.Forst. & G.Forst. and Zanthoxylum L. ( Gregor, Thorne and Kishino, 2002 ) uses a probabilistic model to describe the change in 1989 ). Because the macrofossils can be exclusively attributed to these genera evolutionary rate over time and uses the Markov chain Monte Carlo (MCMC) (Gregor, 1978, 1989 ; Tiffney, 1980; Mai, 1995), both were used as calibration procedure to derive the posterior distribution of rates and time. It allows mul- points for our age estimates. The oldest known records of Euodia and Zan- tiple calibration points and provides direct credibility intervals for estimated thoxylum date from the Late Eocene (35 Ma) and were treated as minimum ages divergence times and substitution rates. We used the programs BEAST 1.4.8 for the stem of the corresponding lineages in the rbcL analysis (Fig. 2). For the and Tracer ( Drummond and Rambaut, 2007 ), which do not assume autocorrela- BEAST analysis, we modeled the clades including Euodia and Zanthoxylum , tion, and MULTIDIVTIME ( Thorne et al., 1998 ; Kishino et al., 2001 , Thorne each, as an exponential distribution ( Ho and Philips, 2009 ) with a mean of 16.4 and Kishino, 2002 ), which does. and an offset of 35 Ma, which corresponds to the maximum age of these fossils. For the BEAST analysis of the rbcL data set, the model parameters deter- With this mean value, the 95% distribution covered 84 Ma, which was regarded mined as optimal by AIC (see results) under the GTR+ +I model and sug- as big time in the MULTIDIVTIME analysis. gested priors taken from a prerun analysis were used. A relaxed clock model Because rbcL was availabile only for a limited number of species, we used with an uncorrelated log-normal rate change was chosen. We tuned the opera- the age estimates for some nodes of the rbcL tree for a BEAST analysis of the tors using BEAST’ s auto-optimization option. We then executed two runs of combined ITS/ matK / atpB-rbcL spacer data set including all Thamnosma spe- 10 7 generations each, sampling every 103 generations, using random starting cies. This procedure followed the BEAST analysis for rbcL with the following trees, and setting the coalescent process and a speciation model following a specifi cations. Mean age estimates with standard deviations of the rbcL BEAST Yule process as tree prior. For all BEAST analyses, resulting posterior distribu- analysis (nodes b and d in Table 3) served as calibration points under a normal   tions for parameter estimates were checked in Tracer 1.4.1 (Drummond and distribution. The GTR+ +I model was chosen. We then executed two runs of 8 4 Rambaut, 2007 ), and maximum credibility trees, representing the maximum a 10 generations each and sampled every 10 generations. posteriori topology, were calculated after removing burn-in with the program TreeAnnotator version 1.4.7. The .xml fi les are available for the Rutaceae anal- Biogeographic analyses— We defi ned the following areas of endemism for ysis as Appendix S1 and for the Thamnosma analysis as Appendix S2 (online at Thamnosma: N = North/Central America, H = Horn of Africa including Socotra http://www.amjbot.org/cgi/content/full/ajb.1000339/DC1). and Somalia, and S = Southern Africa (Table 1). The outgroup, Ruta chalepensis , For the MULTIDIVTIME analyses, we followed the procedure outlined by was coded as Mediterranean, and we did not take into account the potential Thiv et al. (2006) and a step-by-step manual by Rutschmann (2004) . Model Asian relatives (see Materials and Methods: Taxon sampling). To reconstruct parameters for the F84+   model (Kishino and Hasegawa, 1989) were estimated the geographical evolution of Thamnosma , we conducted a dispersal— vicari- using the module BASEML in the program PAML ( Yang, 1997 ). We estimated ance analysis (Ronquist, 1997) using the program S-DIVA 1.9 (Yu et al., 2010). the maximum likelihood of the branch lengths of the rooted evolutionary tree This method calculates the optimized areas over a set of trees, thus taking into together with a variance – covariance matrix of the branch length estimates by account topological uncertainty. We used the 9000 trees retained from the BEAST using the program ESTBRANCHES (Thorne et al., 1998). We used MULTIDI- analysis of the combined data set. The number of maximum areas was set to VTIME to approximate the posterior distributions of substitution rates and three because this refl ects the ingroup’ s number of defi ned areas of distribu- divergence times by using a multivariate normal distribution of estimated tion. To take into account the estimated time between speciation events, we also branch lengths (provided here by ESTBRANCHES) and running the MCMC used the program Lagrange 2.0.1 (Ree and Smith, 2008) for the reconstruction of 80 American Journal of Botany [Vol. 98 ancestral areas, using an ultrametric tree combining the ML topology with American lineages are only weakly supported in all analy- internal node age estimates from a BEAST analysis based on the combined ses, showing bootstrap values (BS) or posterior probabilities data set (not shown). All combinations of areas were allowed in the adjacency (PP) < 53%. matrix, and baseline rates of dispersal and local extinction were estimated. The alternative hypotheses of all northern American species forming a clade and of T. socotrana - T. somalensis being sister to each other resulted each in a difference of tree length of two RESULTS steps in the parametric bootstrap analysis. While the fi rst alter- native cannot be rejected at P > 0.05, the second one is dis- Rutaceae phylogeny— The aligned sequence lengths of rbcL missed at P = 0.04. were 811 bp. The optimal model of sequence evolution for this data set was the transversion (TVM+I+ ) model: unequal base Molecular dating — Enforcing the molecular clock resulted frequencies (A = 0.2721, C = 0.2039, G = 0.2442, T = 0.2798), in a log-likelihood score of − lnL = 2989.19 for the rbcL data six substitution types (A/C: 1.5287, A/G: 2.7801, A/T: 0.5796, set. The comparison between clock and nonclock (LR = C/G: 0.9464, C/T: 2.7801), gamma distribution of rates among − 3356.16) trees by applying the likelihood ratio (LR) test did sites with alpha shape parameter = 0.7607, and the proportion not reject clock-like evolution for this data set (df = 36, P < of invariable sites = 0.6326. The analysis using these parame- 0.05). The results of the two runs using BEAST were very simi- ters yielded a ML tree with a log-likelihood score of – lnL = lar and were therefore combined. The same applied to the two 2962.76. According to the ML tree ( Fig. 2 ), Thamnosma is runs using MULTIDIVTIME. The estimated mean ages and highly supported as monophyletic (99% bootstrap value), and 95% highest posterior density intervals (HPD) are shown in the position as sister to the Mediterranean Ruta is supported by Table 3 referring to nodes of a clock-like ML tree in Fig. 2. For a 75% bootstrap value. In the rbcL study, both genera appear the rbcL data set, the mean substitution rate as indicated by weakly supported as sister to the Californian Cneoridium . Inter- Tracer is 2.96   10 − 10 substitution · site − 1 · yr − 1 . The ages of the specifi c relationships of Thamnosma inferred from a reduced dated lineages inferred by BEAST are generally older than taxon sample were partly resolved by rbcL data. Except for the those by MULTIDIVTIME. Nonetheless, there is a broad over- position of T. trifoliata, they correspond to the results of the lap in HPD between the outcomes of the two programs. The second, more detailed analysis reported next. consensus tree of the BEAST analysis of all Thamnosma spe- cies is shown in Fig. 4. Most important for our considerations Thamnosma phylogeny— The aligned sequence lengths are the following nodes ( Table 3 ). The largest divergence is were 224 bp (ITS1), 231 bp (ITS2), 664 bp ( matK), and 994 bp found for of the stem group of Thamnosma with mean values of ( atpB - rbcL spacer), resulting in a total length of 2113 bp, of 36.75 (14.66 – 61.55) Ma using BEAST and 22.25 (9.21 – 34.92) which 0.89% were scored as missing data. Of these characters, Ma as calculated by MULTIDIVTIME. The mean estimated 248 were variable and 62 parsimony informative. A partition age of the crown node of Thamnosma is 14.56 Ma (4.78– 26.35) homogeneity test explained incongruence between the nuclear and 13.11 (3.33 – 26.83), respectively. The average age of the and chloroplast data at P = 0.954. Therefore, we combined the split between New and Old World Thamnosma was calculated data sets. The MP analysis yielded a single most parsimonious as 12.63 (8.37– 14.94, node F in Table 3) Ma. The age of the tree with a length of 306 steps, a CI of 0.92 and a RI of 0.80. southern African and Horn of Africa clade is 8.53 (5.28– 12.11, The optimal model of sequence evolution for this combined node E in Table 3 ) Ma. Divergence between Socotran T. so- data set was the general time reversible (GTR+I+ ) model: un- cotrana and east African-Arabian(-Socotran) T. hirschii was equal base frequencies (A = 0.2793, C = 0.2042, G = 0.2169, T = determined to be 5.15 (0.36– 11.38) and 4.09 (0.13– 13.65, node 0.2996) and six substitution types (A/C: 1.0607, A/G: 1.4094, D/d in Table 3 ) Ma. A/T: 0.3057, C/G: 1.3575, C/T: 3.0006), gamma distribution of rates among sites with alpha shape parameter = 0.6826, and the Biogeography— The results of Lagrange and DIVA analyses proportion of invariable sites = 0.5533. The analysis using these are given in Table 4 in which nodes refer to those in Figs. 3 and parameters yielded a ML tree with a log-likelihood score of 4. Using the the ML tree, we optimized the two most basal – lnL = 4667.27. nodes (B and G) with the splits N|NH, N|NS, and N|N in The MP, ML, and Bayesian analyses, all indicate that the Lagrange and mostly with N as ancestral area in S-DIVA. The North American species are paraphyletic relative to the Old lineage of T. montana and the Old World yielded almost equally World taxa (Figs. 3, 4). All analyses recognize a North Ameri- likely N|H and N|S, and NH, NS, and NHS. Within the palaeo- can clade consisting of T. stanfordii , T. pailensis, and T. texana tropical clade, Lagrange and DIVA clearly indicate a split be- and a palaeotropical lineage including T. africana , T. crenata , tween S and H. T. rhodesica , T. somalensis , T. hirschii, and T. socotrana . This latter clade is subdivided into the southern African T. africana- T. crenata -T. rhodesica, and the Eastern African-Arabian- DISCUSSION Socotran T. socotrana - T. somalensis - T. hirschii. The results differ with regard to the positions of T. trifoliata and T. mon- Phylogeny of Thamnosma— Our rbcL data support the tana . In the MP study, T. montana is sister to the rest of the genus, monophyly of Thamnosma and its close relationship to Ruta and and T. trifoliata appears as sister to T. stanfordii -T. pailen- Cneoridium , thus corroborating the results of Salvo et al. (2008). sis -T. texana. The ML tree places T. trifoliata at the base, fol- The palaeotropical species of Thamnosma form a strongly sup- lowed by T. stanfordii- T. pailensis- T. texana and T. montana ported clade, which contains two subclades: a southern African and as sister to the Old World clade. The Bayesian analysis does a eastern African-Arabian-Socotran subclade. Beside high BS not resolve relationships between T. trifoliata , T. stanfordii - and PP values, the occurrence of segmented leaves also supports T. pailensis - T. texana , but also indicates T. montana is sister to the southern African clade. No morphological synapomorphy the palaeotropical group. The relationships between these North could be found for the eastern African-Arabian-Socotran clade. January 2011] Thiv et al. —T HAMNOSMA 81

Fig. 2. Clock-like maximum likelihood (ML) tree of selected Rutaceae based on rbcL sequences. Capitals A and B indicate calibration points. Lowercase letters a – d refer to nodes in Table 3 . ML bootstrap values ( > 50%) are on the branches. Numbers behind taxon names refer to EMBL accession numbers ( Table 2 ).

The three species differ mainly in growth form and in quantita- more, the monophyly of the North American species cannot be tive traits like petal, fruit, and seed lengths. The sister-group ruled out by parametric bootstrap analysis. The only well-sup- relationship between T. somalensis and T. hirschii is supported ported relationship is the sister-species relationship between by high bootstrap and PP values. The alternative relationship of T. pailensis and T. texana , which is morphologically corrobo- T. somalensis as sister to T. socotrana, proposed by Thulin rated by the possession of yellow petals ( Johnston, 1983 ). The (1999) , is not corroborated by our data and is rejected by the different positions of T. trifoliata and T. montana might hint at parametric bootstrap analysis. The palaeotropical clade appears low DNA divergence, ancient hybridization events or long branch to be nested within North American species. The phylogenetic attraction. Here we assume that the weakly supported paraphyly relationships among the North American species remain ambigu- of the North American Thamnosma species relative to the Old ous, and no single topology is well supported ( Fig. 3 ). Further- World ones is correct. 82 American Journal of Botany [Vol. 98

Fig. 3. Phylogenetic reconstructions of Thamnosma based on nrITS, matK , and atpB-rbcL intergenic spacer sequences. Left is shown the almost identical cladogram of the Bayesian and maximum likelihood (ML) analyses. ML resolves the positions of T. trifoliata and T. stanfordii -T. pailensis / T. texana as indicated by the bracked line. Posterior probabilities and ML bootstrap values (> 50%) are in this order on the branches. Single MP tree is shown on the right including bootstrap values. For detailed species distribution, see Table 1. T. = Thamnosma. Numbers behind species names refer to collection numbers as shown in Table 2 . Capitals B and D– G refer to nodes in Tables 3 and 4 .

Is Thamnosma a member of the Madro-Tertiary geo- geofl ora. Alternatively, Thamnosma may have evolved outside fl ora?— Axelrod (1958) described a Madro-Tertiary geofl ora, the Madro-Tertiary geofl ora and later migrated into its modern which was widespread in North America since the Eocene – distribution range in the southern United States and Mexico. If Oligocene boundary (34 Ma). Members of this fl ora were char- the present North American Thamnosma are relictual elements acterized by sclerophyllous, microphyllous leaves, which also of the Madro-Tertiary geofl ora, then we predict that the time of suggested an arid, warm climate. Despite the fact that the leaves radiation of North American Thamnosma coincides with the of Thamnosma are nanophyllous and deciduous, the ecology is period of the Madro-Tertiary geofl ora, about 34 – 2 Ma. still consistent with it being a member of the Madro-Tertiary According to our molecular clock calculations, the stem geofl ora. The present distribution of the North American Tham- group age of Thamnosma is either about 22 Ma (9.2 – 34.9) using nosma in southwestern North America corresponds largely with MULTIDIVTIME or 36.8 Ma (14.7– 61.6) using BEAST. These the center of the Madro-Tertiary geofl ora. All these taxa grow ages may be overestimated because Boenninghausenia and Psi- in arid-adapted vegetation ( Johnston, 1924 , 1943 , 1983 ). Fur- lopeganum may break up this branch. Both age estimates of ther evidence comes from the association of extant Thamnosma Thamnosma correspond to the time when the Madro-Tertiary with taxa listed by Axelrod (1958) as elements of the Madro- geofl ora was expanding in the Miocene (Axelrod, 1958; Gra ham, Tertiary geofl ora, suggesting that Thamnosma, despite the ab- 1999). The radiation of the North American Thamnosma started sence of fossils, may have been part of the Madro-Tertiary around 13– 15 (3.3– 26.8) Ma (node b in Fig. 2, Table 3). Thus, our January 2011] Thiv et al. —T HAMNOSMA 83

Fig. 4. Chronogram of Thamnosma of the Bayesian dating analysis using BEAST. Posterior probability values are shown at nodes. Gray bars indicate 95% highest posterior density (HPD) of age estimates. Other features as in Fig. 3 . results are consistent with an early radiation of Thamnosma in the tween the North American and Asian Eocene fl oras ( Wolfe, 1975 ; Madro-Tertiary geofl ora, i.e., the ancestors of the extant species Lavin and Luckow, 1993 ), possibly via the Beringian land bridge. may have been part of this fl ora. If the east Asian Psilopeganum and/or Boenninghausenia are the closest relatives of Thamnosma , New – Old World disjunctions in the succulent biome an origin of the North American species from the Neotropics or an fl ora— Our data enable us to test two hypotheses on the origin Arcto-Tertiary stock ( Axelrod, 1958 ), as suggested for Rhus L. of North American and Eurasian– African disjunctions of taxa (Anacardiaceae) by molecular data (Miller et al., 2001), can be of the succulent biome. First, vicariance hypotheses predict that ruled out. More plausible are connections to tropical to subtropical the age of the disjunction should match the age of the land con- Asia supporting the boreotropics hypothesis postulating links be- nections between the two continents: these may even be sorted 84 American Journal of Botany [Vol. 98

Table 3. Results of the Bayesian dating using BEAST and nested in a paraphyletic American assemblage of taxa. Exam- MULTIDIVTIME showing combined mean ages and the 95% highest ples can be found in the legumes (Lavin et al., 2004) and in posterior density (HPD) of two runs. Nodes in lowercase letters refer Acleisanthes A. Gray incl. Selinocarpus A. Gray (Nyctag- to Fig. 2 and those in capital letters to Figs. 3 and 4 . inaceae, Levin, 2000, 2002 ) where a single Somalian species is BEAST MULTIDIVTIME nested within the American clade of 15 species. We regard this explanation as the most likely one for this disjunction. Node Mean (Ma) 95% HPD (Ma) SE Mean (Ma) 95% HPD (Ma) a 36.75 14.66-61.55 0.38 22.25 9.21 – 34.92 African patterns— Thamnosma radiated in southern and b/B 14.56 4.78-26.35 0.16 13.11 3.33 – 26.83 northeast Africa, giving rise to two geographically separated c 10.52 2.80-20.43 1.12 8.60 1.22 – 20.96 clades (node E in Table 4). These six species map a track that d/D 5.15 0.36-11.38 0.06 4.09 0.13 – 13.65 connects the arid regions of southwestern Africa (Namibia, E 8.53 5.28-12.11 0.02 — — F 12.63 8.37-14.94 0.03 — — Botswana, the northern parts of South Africa) with those of northeastern Africa (Kenya, Somalia, Ethiopia). This “ arid cor- ridor ” has been documented for many plant groups ( Balinsky, by the age of the last connection (North Atlantic or Beringia). 1962 ; Verdcourt, 1969 ; de Winter, 1971 ; J ü rgens, 1997 ). The Second, long-distance dispersal is the most likely explanation age of this corridor, specifi cally the time of its initiation and when all vicariance patterns can be rejected. also closure, is disputed. Some estimates date it to the Pleisto- The vicariance hypotheses cannot be so easily rejected. The cene, thus linking it to the glacial – interglacial Pleistocene cycles. estimated age of the split between the American and the African Other estimates suggest a Miocene date and a severance of the taxa in the Miocene (12.63, 8.37– 14.94 Ma, node F in Table 3) corridor due to Late Miocene uplift and associated rifting in falls outside the time frames of both Axelrod ’ s (1975) Madrean- Africa (Chorowicz, 2005) or major climatic changes during the Tethyan (37.2– 23 Ma) and Tiffney ’ s (1985b) Eocene North Late Miocene ( Schuster et al., 2006 ; Bobe, 2006 ). The age of Atlantic land bridge (55.8 – 33.9 Ma) hypotheses. This is incon- the diversifi cation between the southern African and northeast sistent with vicariance across the North Atlantic, and these re- Africa clades of Thamnosma is in the Lower Miocene (8.53 sults are congruent with those of Lavin et al. (2004) for legumes [5.28 – 12.11] Ma, Table 3 ). This coincides with the age estimates and Hohmann et al. (2006) for the Californian – Mediterranean of corresponding lineages of Androcymbium ( Caujape-Castells Aphanisma – Oreobliton (Amaranthaceae) disjunction. More- et al., 2001) and supports Balinsky ’ s (1962) suggestion that the over, the Madrean-Tethyan hypothesis referred to the Mediter- arid corridor existed already before the Pleistocene, not ruling ranean area rather than the succulent biome in the strict sense of out repeated openings and closures in different periods. How- Schrire et al. (2005a). Our age estimates cannot reject vicari- ever, the possibility that the distribution range was established ance across an early Miocene (23– 16 Ma) migration route via by long-distance dispersal, and not by migration through the the Bering land bridge (Stebbins and Day, 1967; Tiffney, 1985a; arid corridor, cannot be discounted by our data. Hohmann et al., 2006). The much larger distance between North Thamnosma socotrana, an endemic of the Haggeher moun- America and Africa via Beringia and the absence of Tham- tains on Socotra ( Miller and Morris, 2004 ), is basal in the east nosma in Asia argue against it (cf. Hohmann et al., 2006 ). It is African clade. Although today geographically closer to the possible, though unlikely, that the succulent biome, with Tham- Horn of Africa, the Socotran archipelago was separated from nosma in it, may have existed undetected in Asia during the the Arabian peninsula 18 – 15 Ma ( Richardson et al., 1995 ; Fleit- Neogene. Consequently a vicariance explanation for the dis- mann et al., 2004 ; Van Damme, 2009 ). Both vicariance and junction is unlikely. long-distance dispersal have been proposed as explanations for Because the vicariance hypotheses are unlikely, we are left the source of the Socotran endemics (Mies, 2001; Miller and with the long-distance dispersal hypothesis. Numerous cases of Morris, 2004 ). A vicariance explanation for the presence of long-distance dispersal from North America to Africa during T. socotrana on the island predicts that the species should have the Miocene have been postulated, based on African clades separated from its nearest continental relatives at least 18 Ma. The Upper Miocene-Pleistocene date (node d/D in Table 3) for Table 4. Results of the biogeographic reconstructions using Lagrange the separation of T. socotrana from the other two species of the and S-DIVA. Abbreviations and nodes refer to Figs. 3 and 4. East African clade clearly falsifi es the vicariance hypothesis. Our estimated date is in accordance with other recent studies Lagrange S-DIVA focusing on Socotra (Lavin et al., 2000; Thulin and Lavin, Relative Ancestral Frequency 2001 ; Thiv et al., 2006 ; 2010 ), showing that most of the fl ora Node Split -lnL probability area (%) arrived by long-distance dispersal. B [N|NH] 16.66 0.41 N 56.4 [N|NS] 16.67 0.40 NS 14.6 Niche conservation— All species of Thamnosma are re- [N|N] 17.47 0.18 NHS 14.5 stricted to dry environments as defi ned by K ö ppen ( Peel et al., NH 14.5 2007 ) and specifi cally to the succulent biome ( Schrire et al., E [S|H] 15.83 0.93 SH 99.9 2005a ). The small (nano- and leptophyllous) deciduous leaves H 0.1 S 0.1 most likely constitute an adaptation to this seasonally arid cli- F [N|H] 16.50 0.48 NHS 34.1 mate. In concordance with our age estimates, the presence of [N|S] 16.51 0.47 NH 33.0 semiarid conditions in North America in the Middle Miocene NS 33.0 ( Axelrod, 1958 ) served as source of a preadapted aridland Old G [N|NH] 16.58 0.44 N 77.8 world lineage. On a local scale, however, there is some differ- [N|NS] 16.59 0.43 NHS 7.4 entiation visible. As discussed by Lavin et al. (2004) and [N|N] 17.85 0.12 NS 7.4 NH 7.4 Pennington et al. (2009) , the arid biome is characterized glob- ally by highly isolated regions, and consequently most species January 2011] Thiv et al. —T HAMNOSMA 85 are narrow endemics. Thamnosma seems to parallel the patterns biogeography of Androcymbium Willd.(Colchicaceae) in Africa: found for legumes. Only T. texana and T. montana are more Evidence from cpDNA RFLPs. Botanical Journal of the Linnean widespread in the southern United States and Central America, Society 136 : 379 – 392 . all other species occur allopatrically in very restricted areas Chorowicz , J. 2005 . The East African Rift system. Journal of African (Table 1). Among the eastern African species, T. hirschii grows Earth Sciences 43 : 379 – 410 . Coleman , M. , A. Liston , J. W. Kadereit , and R. J. Abbott . 2003 . in several vegetation types from semidesert open grass land to Repeat intercontinental dispersal and Pleistocene speciation in dis- evergreen bushland at altitudes between 10– 1500 m a.s.l. The junct Mediterranean and desert Senecio (Asteraceae). American remaining two species occupy much narrower, montane habitats Journal of Botany 90 : 1446 – 1454 . that mainly differ in their substrate. Thamnosma socotrana is Crisp , M. D. , M. T. K. Arroyo , L. G. Cook , M. A. Gandolfo , G. J. found on granite, while T. somalensis is known from limestone Jordan , M. S. McGlone , P. H. Weston , et al . 2009 . Phylogenetic slopes ( Thulin, 1999 ). Some differentiation can also be observed biome conservatism on a global scale. Nature 458 : 754 – 756 . in the southern African lineage. Here, T. africana and T. rhodesica da Silva , M. F. D. G. F. , O. R. Gottlieb , and F. Ehrendorfer . prefer dry bushland and rock crevices, while T. crenata extends 1988 . Chemosystematics of the Rutaceae: Suggestions for a more its habitat to grasslands ( Thulin, 1999). natural taxonomy and evolutionary interpretation of the family. Plant The genus does not penetrate in Africa into the more mesic Systematics and Evolution 161 : 97 – 134 . De Winter , B. 1971 . Floristic relationships between the northern savannas (e.g., the Central African Miombo or the West Afri- and southern arid areas in Africa. Mitteilungen der botanischen can savannas), nor into the rainforests of the Congo basin or the Staatssammlung M ü nchen 10: 424 – 437. East African coast. It is also absent from the cooler and more Donoghue , M. J. 2008 . A phylogenetic perspective on the distribution mesic evergreen forests and temperate grasslands of the Afri- of plant diversity. Proceedings of the National Academy of Sciences, can uplands, which are widespread from Malawi to Ethiopia USA 105 : 11549 – 11555 . along the African rift valley system. Because all species in the Drummond , A. J. , and A. Rambaut. 2007 . BEAST: Bayesian evo- genus share this ecology (the combination of which habitats are lutionary analysis by sampling trees. BMC Evolutionary Biology actually occupied, and from which it is missing), this pattern 7 : 214 . can be interpreted as a phylogenetic niche conservatism. A di- Engler , A. 1897 . Rutaceae. In A. Engler and K. Prantl [eds.], Die rect result of this conservatism is the fragmented distribution in natü rlichen Pfl anzenfamilien, 95 – 201. Wilhelm Engelmann, Leipzig, Germany. Africa. Engler , A. 1931 . Rutaceae. In A. Engler and H. Harms [eds.], Die natü r- lichen Pfl anzenfamilien, 187 – 359. Wilhelm Engelmann, Leipzig, Conclusion— The diversity of Thamnosma may be the result Germany. of three processes: phylogenetic niche conservatism, rare long- Felsenstein , J. 1981 . Evolutionary trees from DNA sequences: A distance dispersal, and local differentiation. The fi rst constrains maximum likelihood approach. Journal of Molecular Evolution 17 : the genus to the succulent biome, the second allows occupation 368 – 376. of these geographically separated regions, without genetically Fleitmann , D. , A. Matter , S. J. Burns , A. Al-Subbary , and M. A. Al- swamping locally differentiating lineages, and the last allows Aowah . 2004 . Geology and Quaternary climate history of Socotra. the evolution of local specialist species. The result is a global Fauna of Arabia 20 : 27 – 43 . succulent biome, with many shared genera or clades but a few Freitag , S. , and T. J. Robinson . 1993 . Phylogeographic patterns in shared species. mitochondrial DNA of the ostrich (Struthio camelus ). Auk 110 : 614 – 622 . Goldman , N. , J. P. Anderson , and A. G. Rodrigo . 2000 . Likelihood- LITERATURE CITED based tests of topologies in phylogenetics. Systematic Biology 49 : 652 – 670 . Axelrod , D. I. 1958 . Evolution of the Madro-Tertiary geofl ora. Botanical Graham , A. 1999 . Late Cretaceous and Cenozoic history of North Review 24 : 433 – 509 . American vegetation. Oxford University Press, New York, New Axelrod , D. I. 1972 . Edaphic aridity as a factor in angiosperm evolution. York, USA. American Naturalist 106 : 311 – 320 . Gregor , H. J. 1978 . Subtropische Elemente im europ ä ischen Terti ä r Axelrod , D. I. 1973 . History of the Mediterranean ecosystem in III Rutaceae Die Gattungen Toddalia und Zanthoxylum. Acta California. In F. di Castri and H. A. Mooney [eds.], Mediterranean Palaeobotanica 19 : 21 – 40 . type ecosystems— Origin and structure, 225– 277. Springer Verlag, Gregor , H. J. 1989 . Aspects of the fossil record and phylogeny of the Berlin, Germany. family Rutaceae (Zanthoxyleae, Toddalioideae). Plant Systematics Axelrod , D. I. 1975 . Evolution and biogeography of Madrean-Tethyan and Evolution 162 : 251 – 265 . sclerophyll vegetation. Annals of the Missouri Botanical Garden 62 : Ho , S. Y. W. , and M. J. Philips . 2009 . Accounting for calibration un- 280 – 334 . certainty in phylogenetic estimation of evolutionary divergence times. Balinsky , B. I. 1962 . Patterns of animal distribution on the African con- Systematic Biology 58 : 367 – 380 . tinent. Annals of the Cape Provincial Museum 2 : 299 – 310 . Hohmann , S. , J. W. Kadereit , and G. Kadereit . 2006 . Understanding Bellstedt , D. U. , L. van Zijl , E. M. Marais , B. Bytebier , C. A. Mediterranean– Californian disjunctions: Molecular evidence from de Villiers , A. M. Makwarela , and L. L. Dreyer . 2008 . Chenopodiaceae-Betoideae. Taxon 55 : 67 – 78 . Phylogenetic relationships, character evolution and biogeography of Huelsenbeck , J. P. , D. M. Hillis , and R. Jones . 1996 . Parametric southern African members of Zygophyllum (Zygophyllaceae) based bootstrapping in molecular phylogenies: Application and perfor- on three plastid regions. Molecular Phylogenetics and Evolution 47 : mance. In J. D. Ferris and S. R. Palumbi [eds.], Molecular zoology: 932 – 949 . Advances, strategies, and protocols, 19 – 45. Wiley, New York, Blattner , F. R. 1999 . Direct amplifi cation of the entire ITS region New York, USA. from poorly preserved plant material using recombinant PCR. Huelsenbeck , J. P. , and F. Ronquist . 2001 . MrBayes: Bayesian infer- BioTechniques 27 : 1180 – 1186 . ence of phylogeny. Bioinformatics 17 : 754 – 755 . Bobe , R. 2006 . The evolution of arid ecosystems in eastern Africa. Johnson , L. A. , and D. E. Soltis . 1998 . Assessing congruence: Empirical Journal of Arid Environments 66 : 564 – 584 . examples from molecular data. In D. E. Soltis, P. S. Soltis, and J. J. Caujap é -Castells , J. , R. K. Jansen , N. Membrives , J. Pedrola- Doyle [eds.], Molecular systematics of plants II: DNA sequencing, Monfort , J. M. Montserrat , and A. Ardanuy . 2001 . Historical 297 – 348. Kluwer, New York, New York, USA. 86 American Journal of Botany [Vol. 98

Johnston , I. M. 1924 . Expedition of the Californian Academy of Rambaut , A. , and N. C. Grassly. 1997 . Seq-Gen: An application Sciences to the Gulf of California in 1921: The botany (the vascular for the Monte Carlo simulation of DNA sequence evolution along plants). Proceedings of the California Academy of Sciences, series 4 phylogenetic trees. Computer Applications in the Biosciences 1 3 : 12 : 951 – 1218 . 235 – 238 . Johnston , I. M. 1943 . Noteworthy species from Mexico and adjacent Ree , R. H. , and S. A. Smith . 2008 . Maximum likelihood inference of United States, I. Journal of the Arnold Arboretum 24 : 227 – 236 . geographic range evolution by dispersal, local extinction, and clado- Johnston , M. C. 1983 . Thamnosma pailensis (Rutaceae), new spe- genesis. Systematic Biology 57 : 4 – 14. cies from the Sierra de la Paila, Coahuila Mexico. Phytologia 53 : Richardson , S. M. , W. F. Bott , B. A. Smith , W. D. Hollar , and 179 – 180 . P. M. Bermingham . 1995 . A new hydrocarbon “ play ” area offshore Jü rgens , N. 1997 . Floristic biodiversity and history of African arid re- Socotra Island, Republic of Yemen. Journal of Petroleum Geology gions. Biodiversity and Conservation 6 : 495 – 514 . 77 : 225 – 244 . Kishino , H. , and M. Hasegawa . 1989 . Evaluation of the maximum Ronquist , F. 1997 . Dispersal-vicariance analysis: A new approach to likelihood estimate of the evolutionary tree topologies from DNA the quantifi cation of historical biogeography. Systematic Biology 46 : sequence data, and the branching order in Hominoidea. Journal of 195 – 203 . Molecular Evolution 29 : 170 – 179 . Rutschmann , F. 2004 . Bayesian molecular dating using PAML/ Kishino , H. , J. L. Thorne , and W. J. Bruno . 2001 . Performance of a multidivtime. A step-by-step manual [online]. University of divergence time estimation method under a probabilistic model of rate Zurich, Switzerland. Available at ftp://statgen.ncsu.edu/pub/thorne/ evolution. Molecular Biology and Evolution 18 : 352 – 361 . bayesiandating1.5.pdf [6 December 2010]. Lavin , M. , and M. Luckow . 1993 . Origins and relationships of tropi- Salvo , G. , G. Bacchetta , F. Ghahremaninejad , and E. Conti . 2008 . cal North America in the context of the boreotropics hypothesis. Phylogenetic relationships of Ruteae (Rutaceae): New evidence from American Journal of Botany 80 : 1 – 14 . the chloroplast genome and comparisons with non-molecular data. Lavin , M. , M. Thulin , J. N. Labat , and R. T. Pennington. 2000 . Molecular Phylogenetics and Evolution 49 : 736 – 748 . Africa, the odd man out: Molecular biogeography of dalbergioid legumes Sanderson , M. J. 1998 . Estimating rate and time in molecular phylog- (Fabaceae) suggests otherwise. Systematic Botany 25 : 449 – 467 . enies: Beyond the molecular clock? In D. E. Soltis, P. S. Soltis, and Lavin , M. , B. P. Schrire , G. Lewis , R. T. Pennington , A. Delgado- J. J. Doyle [eds.], Molecular systematics of plants, 242 – 264. Kluwer, Salinas , M. Thulin , C. E. Hughes , et al . 2004 . Metacommunity New York, New York, USA. process rather than continental tectonic history better explains Sang , T. , D. J. Crawford , and T. F. Stuessy . 1997 . Chloroplast geographically structured phylogenies in legumes. Philosophical DNA phylogeny, reticulate evolution, and biogeography of Paeonia Transactions of the Royal Society of London, B, Biological Sciences (Paeoniaceae). American Journal of Botany 84 : 1120 – 1136 . 359 : 1509 – 1522 . Schrire , B. D. , M. Lavin , N. P. Barker , and F. Forest . 2009 . Levin , R. A. 2000 . Phylogenetic relationships within Nyctaginaceae Phylogeny of the tribe Indigofereae (Leguminosae – Papilionoideae): Tribe Nyctagineae: Evidence from nuclear and chloroplast genomes. Geographically structured more in succulent-rich and temperate set- Systematic Botany 25 : 738 – 750 . tings than in grass-rich environments. American Journal of Botany Levin , R. A. 2002 . Taxonomic status of Acleisanthes, Selinocarpus, and 96 : 816 – 852 . Ammocodon (Nyctaginaceae). Novon 12 : 58 – 63 . Schrire , B. D. , M. Lavin , and G. P. Lewis . 2005a . Global distribu- Macey , J. R. , J. V. Kuehl , A. Larson , M. D. Robinson , I. H. Ugurtas , tion patterns of the Leguminosae: insights from recent phylogenies. N. B. Ananjeva , H. Rahman , et al. 2008 . Socotra Island the forgot- In I. Friis and H. Balslev [eds.], Plant diversity and complexity pat- ten fragment of Gondwana: Unmasking chameleon lizard history with terns: Local, regional and global dimensions, 375– 422. Royal Danish complete mitochondrial-genomic data. Molecular Phylogenetics and Academy of Sciences and Letters, Copenhagen, Denmark. Evolution 49 : 1015 – 1018 . Schrire , B. D. , G. Lewis , and M. Lavin . 2005b . Biogeography of the Mai , D. H. 1995 . Terti ä re Vegetationsgeschichte Europas — Methoden Leguminosae. In G. Lewis, B. D. Schrire, B. Mackinder, and J. M. und Ergebnisse. Gustav Fischer Verlag, Jena, Germany. Lock [eds.], Legumes of the world, 21 – 54. Royal Botanic Gardens, Manen , J.-F. , A. Natali , and F. Ehrendorfer . 1994 . Phylogeny of Kew, UK. Rubiaceae-Rubieae inferred from the sequence of a cpDNA intergene Schuster , M. , P. Duringer , J. F. Ghienne , P. Vignaud , H. T. Mackaye , region. Plant Systematics and Evolution 190 : 195 – 211 . A. Likius , and M. Brunet . 2006 . The age of the Sahara desert. Mies , B. A. 2001 . Flora und Vegetations ö kologie der Insel Soqotra. Science 311 : 821. Westarp Wissenschaften, Hohenwarsleben, Germany. Song , W.-H. , X.-D. Li , X.-W. Li , H.-W. Huang , and J.-Q. Li . 2004 . Miller , A. G. , and M. Morris. 2004 . Ethnofl ora of the Soqotra Genetic diversity and conservation strategy of Psilopeganum sinense , Archipelago. RBG Edinburgh, Edinburgh, UK. a rare species in the three-gorges reservoir area. Biodiversity Science Miller , A. J. , D. A. Young , and J. Wen. 2001 . Phylogeny and bio- 12 : 227 – 236 . geography of Rhus (Anacardiaceae) based on ITS sequence data. Stebbins , G. L. , and A. Day . 1967 . Cytogenetic evidence for long con- International Journal of Plant Sciences 162 : 1401 – 1407 . tinued stability in the genus Plantago. Evolution 21 : 409 – 428 . Mooney , H. A. , and E. L. Dunn . 1970 . Convergent evolution of Stefanovi 3 , S. , and R. G. Olmstead . 2004 . Testing the phylogenetic Mediterranean-climate evergreen sclerophyll shrubs. Evolution 2 4 : position of a parasitic plant (Cuscuta , Convolvulaceae, Asteridae): 292 – 303 . Bayesian inference and the parametric bootstrap on data drawn from Nei , M. , and S. Kumar . 2000 . Molecular evolution and phylogenetics. three genomes. Systematic Biology 53 : 384 – 399 . Oxford University Press, Oxford, UK. Swofford , D. L. 1998 . PAUP*: Phylogenetic analysis using parsimony Nylander , J. A. A. 2004 . MrModeltest v2. Program distributed by the (*and other methods), version 4.0. Sinauer, Sunderland, Massachusetts, author, Evolutionary Biology Centre, Uppsala University, Uppsala, USA. Sweden. Takhtajan , A. 1986 . Floristic regions of the world. University of Peel , M. C. , B. L. Finlayson , and T. A. Mcmahon . 2007 . Updated world California Press, Berkeley, California, USA. map of the Kö ppen-Geiger climate classifi cation. Hydrology and Thiv , M. , and U. Meve . 2007 . A phylogenetic study of Echidnopsis Earth System Sciences 11 : 1633 – 1644 . Hook. f. (Apocynaceae-Asclepiadoideae): Taxonomic implications Pennington , R. T. , M. Lavin , and A. Oliveira-Filho . 2009 . Woody and the colonization of the Socotran archipelago. Plant Systematics plant diversity, evolution, and ecology in the tropics: Perspectives and Evolution 265 : 71 – 86 . from seasonally dry tropical forests. Annual Review of Ecology Thiv , M. , M. Thulin , M. Hjertson , M. Kropf , and H. P. Linder . 2010 . Evolution and Systematics 40 : 437 – 457 . Evidence for a vicariant origin of Macaronesian– Eritreo/Arabian Posada , D. , and K. A. Crandall . 1998 . Modeltest: Testing the model disjunctions in Campylanthus Roth (Plantaginaceae). Molecular of DNA substitution. Bioinformatics 14 : 817 – 818 . Phylogenetics and Evolution 54 : 607 – 616 . January 2011] Thiv et al. —T HAMNOSMA 87

Thiv , M. , M. Thulin , N. Kilian , and H. P. Linder. 2006 . Eritreo- Arabian Van Damme , K. 2009 . Socotra archipelago. In R. G. Gillespie and affi nities of the Socotran fl ora as revealed from the molecular phylog- D. A. Clague [eds.], Encyclopedia of islands, 846– 851. University of eny of Aerva (Amaranthaceae). Systematic Botany 31 : 560 – 570 . California Press, Berkeley, California, USA. Thompson , J. D. , T. J. Gibson , F. Plewniak , F. Jeanmougin , and Verdcourt , B. 1969 . The arid corridor between the northeast and D. G. Higgins . 1997 . The ClustalX windows interface: Flexible strat- southwest areas of Africa. Palaeoecology of Africa 4 : 140 – 144 . egies for multiple sequence alignment aided by quality analysis tools. Von Balthazar , M. , P. K. Endress , and Y.-L. Qiu . 2000 . Molecular Nucleic Acids Research 25 : 4876 – 4882 . phylogenetics of Buxaceae based on nuclear ITS and plastid ndhF Thorne , J. L. , and H. Kishino. 2002 . Divergence time and evolutionary sequences. International Journal of Plant Sciences 161 : 785 – 792 . rate estimation with multilocus data. Systematic Biology 51 : 689 – 702 . Wiens , J. J. 1998 . Combining data sets with different phylogenetic histo- Thorne , J. L. , H. Kishino , and I. S. Painter . 1998 . Estimating the rate ries. Systematic Biology 47 : 568 – 581 . of evolution of the rate of molecular evolution. Molecular Biology Wiens , J. J. 2004 . Speciation and ecology revisited: Phylogenetic niche and Evolution 15 : 1647 – 1657 . conservatism and the origin of species. Evolution; International Thulin , M. 1999 . Thamnosma (Rutaceae) in Africa. Nordic Journal of Journal of Organic Evolution 58 : 193 – 197 . Botany 19 : 5 – 11 . Wikstr ö m , N. , V. Savolainen , and M. W. Chase . 2001 . Evolution of Thulin , M. , and M. Lavin . 2001 . Phylogeny and biogeography of the angiosperms: Calibrating the family tree. Proceedings of the Royal Ormocarpum group (Fabaceae): A new genus Zygocarpum from the Society of London, B, Biological Sciences 268 : 2211 – 2220 . Horn of Africa region. Systematic Botany 26 : 299 – 317 . Wolfe , J. A. 1975 . Some aspects of plant geography of the northern Tiffney , B. H. 1980 . Fruits and seeds of the Brandon Lignite, V. hemisphere during the late Cretaceous and Tertiary. Annals of the Rutaceae. Journal of the Arnold Arboretum 61 : 1 – 40 . Missouri Botanical Garden 62 : 264 – 279 . Tiffney , B. H. 1985a . Perspectives on the origin of the Floristic similar- Yang , Z. 1997 . PAML: A program package for phylogenetic analysis ity between eastern Asia and eastern North America. Journal of the by maximum likelihood. Computer Applications in the Biosciences Arnold Arboretum 66 : 73 – 94 . 13: 555 – 556 . Tiffney , B. H. 1985b . The Eocene North Atlantic land bridge and its Yu , Y. , A. J. Harris , and X. J. He . 2010 . S-DIVA (Statistical disper- importance in Tertiary and modern phytogeography of the Northern sal – vicariance analysis): A tool for inferring biogeographic histories. Hemisphere. Journal of the Arnold Arboretum 66 : 243 – 273 . Molecular Phylogenetics and Evolution 56: 848 – 850.